DOCSLIB.ORG

A Case Study: Edge Detection Techniques Using Hough Transform and Canny Edge Algorithm

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
A Case Study: Edge Detection Techniques Using Hough Transform and Canny Edge Algorithm International Journal of Mechanical Engineering and Technology (IJMET) Volume 8, Issue 10, October 2017, pp. 729–740, Article ID: IJMET_08_10_079 Available online at http://iaeme.com/Home/issue/IJMET?Volume=8&Issue=10 ISSN Print: 0976-6340 and ISSN Online: 0976-6359 © IAEME Publication Scopus Indexed A CASE STUDY: EDGE DETECTION TECHNIQUES USING HOUGH TRANSFORM AND CANNY EDGE ALGORITHM Deepika Rani Sona, Prayline Rajabai. C, Debopam Dey, Akanksha Jain School of Electronics Engineering, VIT University, Vellore, Tamil Nadu, India Rashmi Ranjan Das School of Electrical Engineering, VIT University, Vellore, Tamil Nadu, India Stephen Olatunde Olabiyisi Department of Computer Science and Engineering, Ladoke Akintola University of Technology, Ogbomoso, Oyo State, Nigeria ABSTRACT Edge Detection is an ever-expanding frontier in the field of Digital Image Processing. Several research papers have been published in this domain in the past decade specifying many applications of edge detection methods using platforms like MATLAB or OpenCV like Hough, Canny, Sobel, Prewitt Trans- forms among many others. In this case study, various elaborations on basic methods of edge detection have been made and their applications have been discussed in real-time. Furthermore, the performance analysis is figure out by comparison between the results of Canny Edge Detection method and Hough Transforms. Keywords: Edge Detection, Canny Edge Detection Algorithm, Computation Time, Hough Transform, Memory Mapping. Cite this Article: Deepika Rani Sona, Prayline Rajabai. C, Debopam Dey, Akanksha Jain, Rashmi Ranjan Das and Stephen Olatunde Olabiyisi, A Case Study: Edge Detection Techniques Using Hough Transform and Canny Edge Algorithm, International Journal of Mechanical Engineering and Technology 8(10), 2017, pp. 729–740. http://iaeme.com/Home/issue/IJMET?Volume=8&Issue=10 1. INTRODUCTION Paul Hough suggested an efficient method for detecting lines and edges in binary images in 1962 [1]. Hough Transform (hereinafter referred as HT) converts a global detection problem in the image space into an easier local peak detection problem in the parameter space. The working algorithm of HT is explained in fig 1 which uses the slope-intercept parameterization and the voting scheme. http://iaeme.com/Home/journal/IJMET 729 [email protected] Deepika Rani Sona, Prayline Rajabai. C, Debopam Dey, Akanksha Jain, Rashmi Ranjan Das and Stephen Olatunde Olabiyisi 1) Building the parameter space with an appropriate quantization level for any given line of slope m and intercept c. 2) Create an accumulator array denoted as A (m, c). 3) Set A (m, c) = 0 ∀ (m, c) 4) Extract the edge of the images using Canny Detector. 5) For each pixel on the image edges: (xi, yi )∀ (mk , cl ) Verifying equation: cl = −ximk + yi ; Increment: A(mk , cl ) = A(mk , cl ) + 1 6) Find the local maxima A(m, c) and in that indicate the lines in the parameter space. For the above algorithm, all edge pixels lying on a line plays a role in increasing the content of the cell in the accumulator array which corresponds to the real slope and intercept of that line. This suggests that every corresponding straight line does really exist in the image. Thus, we can say that the HT can be considered as a vote-counting-procedure where each edge point votes for all the parameter combinations which could have possibly resulted in its formation. Fig. 1 represents a straight line with slope 1 and intercepts 3 and the corresponding pixel values of the image formed by it. Figure 1 A line in image space is represented as a cell: (m=1, c=3) in parameter (Hough) space. Figure 2 shows various images and cases for point−to−curve transformation. Here, we obtain degenerate solutions when the input is a vertical line or nearly vertical line. Thus, the slope approaches to infinity. This is the stimulus which suggests the (ρ,θ) parameterization in [2]. The parameter ρ is the perpendicular distance from the origin to the line, and the parameter θ is the angle between the perpendicular and the horizontal axis. In fact, this is a point-to-curve transformation using the normal representation of the line: xcosθ +ysinθ =ρ All the edge points are independently treated in HT; which makes the parallel processing of all points possible which leads to its real-time applications. Also, all the partially deformed and noisy shapes can be handled very easily due to its voting scheme. Moreover, the multiple occurrences of lines can be detected very easily since each occurrence has its unique cell address in the parameter space. HT can be extended to detect shapes other than lines. All the above-mentioned points give HT an edge over any other edge detection methods. Figure 2 A point-to-curve transformation in (ρ,θ) parameterization http://iaeme.com/Home/journal/IJMET 730 [email protected] A Case Study: Edge Detection Techniques Using Hough Transform and Canny Edge Algorithm Moreover, the HT has certain drawbacks. It requires a high computational demand and a large storage area. However, it can be increased by reducing the size of the accumulator array. Another edge detection technique is Canny Edge Detection Algorithm (hereinafter referred as CEDA) which was developed by John F. Canny in 1986. According to Canny, an optimal edge detection filter should maximize the product of localization criteria and detection with the restriction in low spurious inputs. The main advantage of canny edge detection method is that the algorithm is adaptable to various environments. Its parameters allow it to be modified to the edges of differing characteristics and figures depending on the requirements from the end user in any given implementation or application. The remaining paper is arranged and organized as follows: Section II gives us the methodology followed by both the edge detection techniques separately. Section III contains the shape detection aspects of any image. Section IV presents the speedup and memory saving techniques in both the algorithm. Section V deals with the experimental results obtained in MATLAB. Section VI focuses on all possible real-time applications. Section VII has the future work related to this article and Section VIII concludes the article. 2. METHODOLOGY Hough Transform Figure 3 shows the architecture of the proposed circle detection algorithm. The algorithm consists of two modules; an edge detector and a voting processor. First, in the edge detector module, we apply the Ball Detection Algorithm [4] to obtain edges from grayscale images. To efficiently handle edges, we utilize double-buffered internal memory blocks to store the edge list (hereinafter referred as EL). Next, for each edge pixel, the voting processor constructs a circular cone, as described in [5], in Hough space with adaptive circular resolution using pre- calculated trigonometric values based on the standard circle Hough transform. Finally, relatively high voter circles are chosen from the circular cone. To minimize the high-cost numerical resources, the proposed edge detector module is mainly implemented based on the ball detection algorithm [4] for background subtraction. The trigonometric function values are indexed by n × dtθ (θ interval in the trigonometric function table), as depicted in Fig4. Since we use eight sample points in a pixel, more than eight V Es for a pixel may not have a good effect on the weight calculation. Thus, dtθ can be adaptively decided by the following equation: Figure 3 Block Diagram of the Edge Detection System http://iaeme.com/Home/journal/IJMET 731 [email protected] Deepika Rani Sona, Prayline Rajabai. C, Debopam Dey, Akanksha Jain, Rashmi Ranjan Das and Stephen Olatunde Olabiyisi 128 푘 = 푟+1 , 푘 ≥ 1 푑푡휃 = { 2| | 2 1, 푘 < 1 k can be calculated by simple shift operators without complex dividers and multiplexers since the dividing and multiplexing operations in the equation are binary operations. Figure 4 Structure of the Voting Procedure [2] Figure 4 shows the architecture of the proposed voting processor. First, the Edge fetch unit reads the edge information from EL and the quad-vectored edge generator modules in the Rasterizer Unit produce vectored edges in parallel using pre- calculated values obtained from the trigonometric table. Thus, each module processes only 128/ (4.dtθ ) times with the range of 0o ∼ 21.8o, 22.5o ∼ 44.3o, 45.0o ∼ 66.8o, and 67.5o89.3o. The coordinate values between 90.0o and 359.3ocan be simply calculated with addition/subtraction operators from the results of the modules because they are all symmetric by the x-axis or y-axis, or both. Next, render modules set the flag bits based on the edge flag algorithm [5]. They then accumulate and calculate the weight value of each pixel as described in [6]. Canny Edge Detection Algorithm Canny Edge Detection is one of the simplest methods used in shape recognition. This method uses 4 stages for edge detection which are mentioned as follows [3]: • Image smoothing • Calculating edge strength and edge direction. • Directional non-maximum suppression. • Invoking threshold with Hysteresis. The output is an image containing only the edges detected through these processes. http://iaeme.com/Home/journal/IJMET 732 [email protected] A Case Study: Edge Detection Techniques Using Hough Transform and Canny Edge Algorithm Figure 5 Steps involved in Canny Edge Detection Algorithm Image smoothing is performed by convolving the input image with a Gaussian filter. This process is used to reduce the noise within an image or to produce a less pixilated image. The output obtained is a blurred intermediate image. This blurred image is input to the next block to calculate the edge strength and direction of the edges. The final output of this stage is a blurred image obtained from the image smoothing stage which is convolved with a 3x3 Sobel operator. The Sobel operators are used to obtain a gradient image.
Load more

Recommended publications
  • Hough Transform, Descriptors Tammy Riklin Raviv Electrical and Computer Engineering Ben-Gurion University of the Negev Hough Transform
    DIGITAL IMAGE PROCESSING Lecture 7 Hough transform, descriptors Tammy Riklin Raviv Electrical and Computer Engineering Ben-Gurion University of the Negev Hough transform y m x b y m 3 5 3 3 2 2 3 7 11 10 4 3 2 3 1 4 5 2 2 1 0 1 3 3 x b Slide from S. Savarese Hough transform Issues: • Parameter space [m,b] is unbounded. • Vertical lines have infinite gradient. Use a polar representation for the parameter space Hough space r y r q x q x cosq + ysinq = r Slide from S. Savarese Hough Transform Each point votes for a complete family of potential lines: Each pencil of lines sweeps out a sinusoid in Their intersection provides the desired line equation. Hough transform - experiments r q Image features ρ,ϴ model parameter histogram Slide from S. Savarese Hough transform - experiments Noisy data Image features ρ,ϴ model parameter histogram Need to adjust grid size or smooth Slide from S. Savarese Hough transform - experiments Image features ρ,ϴ model parameter histogram Issue: spurious peaks due to uniform noise Slide from S. Savarese Hough Transform Algorithm 1. Image à Canny 2. Canny à Hough votes 3. Hough votes à Edges Find peaks and post-process Hough transform example http://ostatic.com/files/images/ss_hough.jpg Incorporating image gradients • Recall: when we detect an edge point, we also know its gradient direction • But this means that the line is uniquely determined! • Modified Hough transform: for each edge point (x,y) θ = gradient orientation at (x,y) ρ = x cos θ + y sin θ H(θ, ρ) = H(θ, ρ) + 1 end Finding lines using Hough transform
    [Show full text]
  • Scale Invariant Feature Transform (SIFT) Why Do We Care About Matching Features?
    Scale Invariant Feature Transform (SIFT) Why do we care about matching features? • Camera calibration • Stereo • Tracking/SFM • Image moiaicing • Object/activity Recognition • … Objection representation and recognition • Image content is transformed into local feature coordinates that are invariant to translation, rotation, scale, and other imaging parameters • Automatic Mosaicing • http://www.cs.ubc.ca/~mbrown/autostitch/autostitch.html We want invariance!!! • To illumination • To scale • To rotation • To affine • To perspective projection Types of invariance • Illumination Types of invariance • Illumination • Scale Types of invariance • Illumination • Scale • Rotation Types of invariance • Illumination • Scale • Rotation • Affine (view point change) Types of invariance • Illumination • Scale • Rotation • Affine • Full Perspective How to achieve illumination invariance • The easy way (normalized) • Difference based metrics (random tree, Haar, and sift, gradient) How to achieve scale invariance • Pyramids • Scale Space (DOG method) Pyramids – Divide width and height by 2 – Take average of 4 pixels for each pixel (or Gaussian blur with different ) – Repeat until image is tiny – Run filter over each size image and hope its robust How to achieve scale invariance • Scale Space: Difference of Gaussian (DOG) – Take DOG features from differences of these images‐producing the gradient image at different scales. – If the feature is repeatedly present in between Difference of Gaussians, it is Scale Invariant and should be kept. Differences Of Gaussians
    [Show full text]
  • Exploiting Information Theory for Filtering the Kadir Scale-Saliency Detector
    Introduction Method Experiments Conclusions Exploiting Information Theory for Filtering the Kadir Scale-Saliency Detector P. Suau and F. Escolano {pablo,sco}@dccia.ua.es Robot Vision Group University of Alicante, Spain June 7th, 2007 P. Suau and F. Escolano Bayesian filter for the Kadir scale-saliency detector 1 / 21 IBPRIA 2007 Introduction Method Experiments Conclusions Outline 1 Introduction 2 Method Entropy analysis through scale space Bayesian filtering Chernoff Information and threshold estimation Bayesian scale-saliency filtering algorithm Bayesian scale-saliency filtering algorithm 3 Experiments Visual Geometry Group database 4 Conclusions P. Suau and F. Escolano Bayesian filter for the Kadir scale-saliency detector 2 / 21 IBPRIA 2007 Introduction Method Experiments Conclusions Outline 1 Introduction 2 Method Entropy analysis through scale space Bayesian filtering Chernoff Information and threshold estimation Bayesian scale-saliency filtering algorithm Bayesian scale-saliency filtering algorithm 3 Experiments Visual Geometry Group database 4 Conclusions P. Suau and F. Escolano Bayesian filter for the Kadir scale-saliency detector 3 / 21 IBPRIA 2007 Introduction Method Experiments Conclusions Local feature detectors Feature extraction is a basic step in many computer vision tasks Kadir and Brady scale-saliency Salient features over a narrow range of scales Computational bottleneck (all pixels, all scales) Applied to robot global localization → we need real time feature extraction P. Suau and F. Escolano Bayesian filter for the Kadir scale-saliency detector 4 / 21 IBPRIA 2007 Introduction Method Experiments Conclusions Salient features X HD(s, x) = − Pd,s,x log2Pd,s,x d∈D Kadir and Brady algorithm (2001): most salient features between scales smin and smax P.
    [Show full text]
  • Hough Transform 1 Hough Transform
    Hough transform 1 Hough transform The Hough transform ( /ˈhʌf/) is a feature extraction technique used in image analysis, computer vision, and digital image processing.[1] The purpose of the technique is to find imperfect instances of objects within a certain class of shapes by a voting procedure. This voting procedure is carried out in a parameter space, from which object candidates are obtained as local maxima in a so-called accumulator space that is explicitly constructed by the algorithm for computing the Hough transform. The classical Hough transform was concerned with the identification of lines in the image, but later the Hough transform has been extended to identifying positions of arbitrary shapes, most commonly circles or ellipses. The Hough transform as it is universally used today was invented by Richard Duda and Peter Hart in 1972, who called it a "generalized Hough transform"[2] after the related 1962 patent of Paul Hough.[3] The transform was popularized in the computer vision community by Dana H. Ballard through a 1981 journal article titled "Generalizing the Hough transform to detect arbitrary shapes". Theory In automated analysis of digital images, a subproblem often arises of detecting simple shapes, such as straight lines, circles or ellipses. In many cases an edge detector can be used as a pre-processing stage to obtain image points or image pixels that are on the desired curve in the image space. Due to imperfections in either the image data or the edge detector, however, there may be missing points or pixels on the desired curves as well as spatial deviations between the ideal line/circle/ellipse and the noisy edge points as they are obtained from the edge detector.
    [Show full text]
  • Topic: 9 Edge and Line Detection
    V N I E R U S E I T H Y DIA/TOIP T O H F G E R D I N B U Topic: 9 Edge and Line Detection Contents: First Order Differentials • Post Processing of Edge Images • Second Order Differentials. • LoG and DoG filters • Models in Images • Least Square Line Fitting • Cartesian and Polar Hough Transform • Mathemactics of Hough Transform • Implementation and Use of Hough Transform • PTIC D O S G IE R L O P U P P A D E S C P I Edge and Line Detection -1- Semester 1 A S R Y TM H ENT of P V N I E R U S E I T H Y DIA/TOIP T O H F G E R D I N B U Edge Detection The aim of all edge detection techniques is to enhance or mark edges and then detect them. All need some type of High-pass filter, which can be viewed as either First or Second order differ- entials. First Order Differentials: In One-Dimension we have f(x) d f(x) dx d f(x) dx We can then detect the edge by a simple threshold of d f (x) > T Edge dx ⇒ PTIC D O S G IE R L O P U P P A D E S C P I Edge and Line Detection -2- Semester 1 A S R Y TM H ENT of P V N I E R U S E I T H Y DIA/TOIP T O H F G E R D I N B U Edge Detection I but in two-dimensions things are more difficult: ∂ f (x,y) ∂ f (x,y) Vertical Edges Horizontal Edges ∂x → ∂y → But we really want to detect edges in all directions.
    [Show full text]
  • Histogram of Directions by the Structure Tensor
    Histogram of Directions by the Structure Tensor Josef Bigun Stefan M. Karlsson Halmstad University Halmstad University IDE SE-30118 IDE SE-30118 Halmstad, Sweden Halmstad, Sweden [email protected] [email protected] ABSTRACT entity). Also, by using the approach of trying to reduce di- Many low-level features, as well as varying methods of ex- rectionality measures to the structure tensor, insights are to traction and interpretation rely on directionality analysis be gained. This is especially true for the study of the his- (for example the Hough transform, Gabor filters, SIFT de- togram of oriented gradient (HOGs) features (the descriptor scriptors and the structure tensor). The theory of the gra- of the SIFT algorithm[12]). We will present both how these dient based structure tensor (a.k.a. the second moment ma- are very similar to the structure tensor, but also detail how trix) is a very well suited theoretical platform in which to they differ, and in the process present a different algorithm analyze and explain the similarities and connections (indeed for computing them without binning. In this paper, we will often equivalence) of supposedly different methods and fea- limit ourselves to the study of 3 kinds of definitions of di- tures that deal with image directionality. Of special inter- rectionality, and their associated features: 1) the structure est to this study is the SIFT descriptors (histogram of ori- tensor, 2) HOGs , and 3) Gabor filters. The results of relat- ented gradients, HOGs). Our analysis of interrelationships ing the Gabor filters to the tensor have been studied earlier of prominent directionality analysis tools offers the possibil- [3], [9], and so for brevity, more attention will be given to ity of computation of HOGs without binning, in an algo- the HOGs.
    [Show full text]
  • The Hough Transform As a Tool for Image Analysis
    THE HOUGH TRANSFORM AS A TOOL FOR IMAGE ANALYSIS Josep Llad´os Computer Vision Center - Dept. Inform`atica. Universitat Aut`onoma de Barcelona. Computer Vision Master March 2003 Abstract The Hough transform is a widespread technique in image analysis. Its main idea is to transform the image to a parameter space where clusters or particular configurations identify instances of a shape under detection. In this chapter we overview some meaningful Hough-based techniques for shape detection, either parametrized or generalized shapes. We also analyze some approaches based on the straight line Hough transform able to detect particular structural properties in images. Some of the ideas of these approaches will be used in the following chapter to solve one of the goals of the present work. 1 Introduction The Hough transform was first introduced by Paul Hough in 1962 [4] with the aim of detecting alignments in T.V. lines. It became later the basis of a great number of image analysis applications. The Hough transform is mainly used to detect parametric shapes in images. It was first used to detect straight lines and later extended to other parametric models such as circumferences or ellipses, being finally generalized to any parametric shape [1]. The key idea of the Hough transform is that spatially extended patterns are transformed into a parameter space where they can be represented in a spatially compact way. Thus, a difficult global detection problem in the image space is reduced to an easier problem of peak detection in a parameter space. Ü; Ý µ A set of collinear image points ´ can be represented by the equation: ÑÜ Ò =¼ Ý (1) Ò where Ñ and are two parameters, the slope and intercept, which characterize the line.
    [Show full text]
  • A Comprehensive Analysis of Edge Detectors in Fundus Images for Diabetic Retinopathy Diagnosis
    SSRG International Journal of Computer Science and Engineering (SSRG-IJCSE) – Special Issue ICRTETM March 2019 A Comprehensive Analysis of Edge Detectors in Fundus Images for Diabetic Retinopathy Diagnosis J.Aashikathul Zuberiya#1, M.Nagoor Meeral#2, Dr.S.Shajun Nisha #3 #1M.Phil Scholar, PG & Research Department of Computer Science, Sadakathullah Appa College, Tirunelveli, Tamilnadu, India #2M.Phil Scholar, PG & Research Department of Computer Science, Sadakathullah Appa College, Tirunelveli, Tamilnadu, India #3Assistant Professor & Head, PG & Research Department of Computer Science, Sadakathullah Appa College, Tirunelveli, Tamilnadu, India Abstract severe stage. PDR is an advanced stage in which the Diabetic Retinopathy is an eye disorder that fluids sent by the retina for nourishment trigger the affects the people with diabetics. Diabetes for a growth of new blood vessel that are abnormal and prolonged time damages the blood vessels of retina fragile.Initial stage has no vision problem, but with and affects the vision of a person and leads to time and severity of diabetes it may lead to vision Diabetic retinopathy. The retinal fundus images of the loss.DR can be treated in case of early detection. patients are collected are by capturing the eye with [1][2] digital fundus camera. This captures a photograph of the back of the eye. The raw retinal fundus images are hard to process. To enhance some features and to remove unwanted features Preprocessing is used and followed by feature extraction . This article analyses the extraction of retinal boundaries using various edge detection operators such as Canny, Prewitt, Roberts, Sobel, Laplacian of Gaussian, Kirsch compass mask and Robinson compass mask in fundus images.
    [Show full text]
  • Robust Edge Aware Descriptor for Image Matching
    Robust Edge Aware Descriptor for Image Matching Rouzbeh Maani, Sanjay Kalra, Yee-Hong Yang Department of Computing Science, University of Alberta, Edmonton, Canada Abstract. This paper presents a method called Robust Edge Aware De- scriptor (READ) to compute local gradient information. The proposed method measures the similarity of the underlying structure to an edge using the 1D Fourier transform on a set of points located on a circle around a pixel. It is shown that the magnitude and the phase of READ can well represent the magnitude and orientation of the local gradients and present robustness to imaging effects and artifacts. In addition, the proposed method can be efficiently implemented by kernels. Next, we de- fine a robust region descriptor for image matching using the READ gra- dient operator. The presented descriptor uses a novel approach to define support regions by rotation and anisotropical scaling of the original re- gions. The experimental results on the Oxford dataset and on additional datasets with more challenging imaging effects such as motion blur and non-uniform illumination changes show the superiority and robustness of the proposed descriptor to the state-of-the-art descriptors. 1 Introduction Local feature detection and description are among the most important tasks in computer vision. The extracted descriptors are used in a variety of applica- tions such as object recognition and image matching, motion tracking, facial expression recognition, and human action recognition. The whole process can be divided into two major tasks: region detection, and region description. The goal of the first task is to detect regions that are invariant to a class of transforma- tions such as rotation, change of scale, and change of viewpoint.
    [Show full text]
  • Third Harmonic Generation Microscopy Zhang, Z
    VU Research Portal Third harmonic generation microscopy Zhang, Z. 2017 document version Publisher's PDF, also known as Version of record Link to publication in VU Research Portal citation for published version (APA) Zhang, Z. (2017). Third harmonic generation microscopy: Towards automatic diagnosis of brain tumors. General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal ? Take down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. E-mail address: [email protected] Download date: 05. Oct. 2021 Third harmonic generation microscopy: towards automatic diagnosis of brain tumors This thesis was reviewed by: prof.dr. J. Hulshof VU University Amsterdam prof.dr. J. Popp Jena University prof.dr. A.G.J.M. van Leeuwen Academic Medical Center prof.dr. M. van Herk The University of Manchester dr. I.H.M. van Stokkum VU University Amsterdam dr. P. de Witt Hamer VU University Medical Center © Copyright Zhiqing Zhang, 2017 ISBN: 978-94-6295-704-6 Printed in the Netherlands by Proefschriftmaken.
    [Show full text]
  • Lecture 10 Detectors and Descriptors
    This lecture is about detectors and descriptors, which are the basic building blocks for many tasks in 3D vision and recognition. We’ll discuss Lecture 10 some of the properties of detectors and descriptors and walk through examples. Detectors and descriptors • Properties of detectors • Edge detectors • Harris • DoG • Properties of descriptors • SIFT • HOG • Shape context Silvio Savarese Lecture 10 - 16-Feb-15 Previous lectures have been dedicated to characterizing the mapping between 2D images and the 3D world. Now, we’re going to put more focus From the 3D to 2D & vice versa on inferring the visual content in images. P = [x,y,z] p = [x,y] 3D world •Let’s now focus on 2D Image The question we will be asking in this lecture is - how do we represent images? There are many basic ways to do this. We can just characterize How to represent images? them as a collection of pixels with their intensity values. Another, more practical, option is to describe them as a collection of components or features which correspond to “interesting” regions in the image such as corners, edges, and blobs. Each of these regions is characterized by a descriptor which captures the local distribution of certain photometric properties such as intensities, gradients, etc. Feature extraction and description is the first step in many recent vision algorithms. They can be considered as building blocks in many scenarios The big picture… where it is critical to: 1) Fit or estimate a model that describes an image or a portion of it 2) Match or index images 3) Detect objects or actions form images Feature e.g.
    [Show full text]
  • Visualize Yolo
    MSc Artificial Intelligence Master Thesis Open the black box: Visualize Yolo by Peter Heemskerk 11988797 August 18, 2020 36 EC credits autumn 2019 till summer 2020 Supervisor: Dr. Jan-Mark Geusebroek Assessor: Prof. Dr. Theo Gevers University of Amsterdam Contents 1 Introduction 2 1.1 Open the neural network black box . .2 1.2 Project Background . .2 1.2.1 Wet Cooling Towers . .2 1.2.2 The risk of Legionellosis . .2 1.2.3 The project . .3 2 Related work 4 2.1 Object Detection . .4 2.2 Circle detection using Hough transform . .4 2.3 Convolutional Neural Networks (ConvNet) . .5 2.4 You Only Look Once (Yolo) . .5 2.4.1 Yolo version 3 - bounding box and class prediction . .6 2.4.2 Yolo version 3 - object recognition at different scales . .7 2.4.3 Yolo version 3 - network architecture . .7 2.4.4 Feature Pyramid Networks . .7 2.5 The Black Box Explanation problem . .9 2.6 Network Visualization . 10 3 Approach 10 3.1 Aerial imagery dataset . 10 3.2 Yolo version 3 . 12 3.2.1 Pytorch Implementation . 12 3.2.2 Tuning approach . 12 3.3 Evaluation . 12 3.3.1 Training, test and validation sets . 12 3.3.2 Evaluation Metrics . 12 3.4 Network Visualization . 13 3.4.1 Introduction . 13 3.4.2 Grad-CAM . 13 3.4.3 Feature maps . 14 4 Experiment 15 4.1 Results . 15 4.1.1 Hough Transform prediction illustration . 15 4.1.2 Yolo Prediction illustration . 16 4.1.3 Yolo Tuning . 17 4.1.4 Yolo Validation .
    [Show full text]